This is the report of the analysis made for the paper (TITLE) AND AUTHORS. INSERT ABSTRACT
Importing data and filtering out those genes with cpm lesser than 1. We use the filtered.data method in NOISeq package.
countMatrix <- ReadDataFrameFromTsv(file.name.path="../../data/refSEQ_countMatrix.txt")
## ../../data/refSEQ_countMatrix.txt read from disk!
# head(countMatrix)
designMatrix <- ReadDataFrameFromTsv(file.name.path="../../design/all_samples_short_names_noHC7.tsv.csv")
## ../../design/all_samples_short_names_noHC7.tsv.csv read from disk!
# head(designMatrix)
filteredCountsProp <- filterLowCounts(counts.dataframe=countMatrix,
is.normalized=FALSE,
design.dataframe=designMatrix,
cond.col.name="gcondition",
method.type="Proportion")
## features dimensions before normalization: 27179
## Filtering out low count features...
## 14495 features are to be kept for differential expression analysis with filtering method 3
Loading Negative Control Genes to normalize data
library(readxl)
sd.ctrls <- read_excel(path="../../data/controls/Additional File 4 full list of BMC genomics SD&RS2.xlsx", sheet=1)
sd.ctrls <- sd.ctrls[order(sd.ctrls$adj.P.Val),]
sd.neg.ctrls <- sd.ctrls[sd.ctrls$adj.P.Val > 0.9, ]
sd.neg.ctrls <- sd.neg.ctrls$`MGI Symbol`
sd.neg.ctrls <- sd.neg.ctrls[-which(is.na(sd.neg.ctrls))]
int.neg.ctrls <- sd.neg.ctrls
int.neg.ctrls <- unique(int.neg.ctrls)
neg.map <- convertGenesViaMouseDb(gene.list=int.neg.ctrls, fromType="SYMBOL",
"ENTREZID")
# sum(is.na(neg.map$ENTREZID))
neg.ctrls.entrez <- as.character(neg.map$ENTREZID)
ind.ctrls <- which(rownames(filteredCountsProp) %in% neg.ctrls.entrez)
counts.neg.ctrls <- filteredCountsProp[ind.ctrls,]
Loading Positive Control Genes to detect them during the differential expression step.
## sleep deprivation
sd.lit.pos.ctrls <- read_excel("../../data/controls/SD_RS_PosControls_final.xlsx",
sheet=1)
colnames(sd.lit.pos.ctrls) <- sd.lit.pos.ctrls[1,]
sd.lit.pos.ctrls <- sd.lit.pos.ctrls[-1,]
sd.est.pos.ctrls <- read_excel("../../data/controls/SD_RS_PosControls_final.xlsx",
sheet=3)
sd.pos.ctrls <- cbind(sd.est.pos.ctrls$`MGI Symbol`, "est")
sd.pos.ctrls <- rbind(sd.pos.ctrls, cbind(sd.lit.pos.ctrls$Gene, "lit"))
sd.pos.ctrls <- sd.pos.ctrls[-which(duplicated(sd.pos.ctrls[,1])),]
sd.pos.ctrls <- sd.pos.ctrls[-which(is.na(sd.pos.ctrls[,1])),]
Normalizing data with TMM, as implemented in edgeR package, and plotting a PCA and an RLE plot of them.
normPropCountsTmm <- NormalizeData(data.to.normalize=filteredCountsProp,
norm.type="tmm",
design.matrix=designMatrix)
Applying a RUVs method of RUVSeq package on normalized data, in order to adjust the counts for the unwanted variation. And of corse we plot a PCA and an RLE plot on these data.
library(RUVSeq)
neg.ctrl.list <- rownames(counts.neg.ctrls)
groups <- makeGroups(designMatrix$gcondition)
ruvedSExprData <- RUVs(as.matrix(round(normPropCountsTmm)), cIdx=neg.ctrl.list,
scIdx=groups, k=5)
normExprData <- ruvedSExprData$normalizedCounts
ggp1 <- PlotPCAPlotlyFunction(counts.data.frame=log1p(normExprData),
design.matrix=designMatrix, shapeColname="condition",
colorColname="genotype", xPCA="PC1", yPCA="PC2",
plotly.flag=FALSE, show.plot.flag=FALSE, save.plot=FALSE,
prefix.plot=NULL)
## [1] FALSE
# ggp2 <- PlotPCAPlotlyFunction(counts.data.frame=log1p(normExprData),
# design.matrix=designMatrix, shapeColname="condition",
# colorColname="genotype", xPCA="PC2", yPCA="PC3",
# plotly.flag=FALSE, show.plot.flag=FALSE, save.plot=FALSE,
# prefix.plot=NULL)
# ggp3 <- PlotPCAPlotlyFunction(counts.data.frame=log1p(normExprData),
# design.matrix=designMatrix, shapeColname="condition",
# colorColname="genotype", xPCA="PC1", yPCA="PC3",
# plotly.flag=FALSE, show.plot.flag=FALSE, save.plot=FALSE,
# prefix.plot=NULL)
# plotly::subplot(ggp1, ggp2, ggp3, nrows=2, margin = 0.1, titleX=TRUE, titleY=TRUE)
ggplotly(ggp1)
dir.create("plots")
save_plot(filename="plots/PCA.pdf", plot=ggp1)
pal <- RColorBrewer::brewer.pal(9, "Set1")
plotRLE(normExprData, outline=FALSE, col=pal[designMatrix$gcondition])
Making differential expression analysis with edgeR package on four different contrasts.
Here is a brief legend:
padj.thr <- 0.05
venn.padgj.thr <- 0.1
desMat <- cbind(designMatrix, ruvedSExprData$W)
colnames(desMat) <- c(colnames(designMatrix), colnames(ruvedSExprData$W))
# cc <- c("WTSD5 - WTHC5", "S3HC5 - WTHC5",
# "S3SD5 - WTSD5", "S3SD5 - S3HC5")
cc <- c("S3HC5 - WTHC5", "S3SD5 - WTSD5", "S3RS2 - WTRS2")
rescList1 <- applyEdgeR(counts=filteredCountsProp, design.matrix=desMat,
factors.column="gcondition",
weight.columns=c("W_1", "W_2", "W_3", "W_4", "W_5"),
contrasts=cc, useIntercept=FALSE, p.threshold=1,
is.normalized=FALSE, verbose=TRUE)
names <- names(rescList1)
rescList1 <- lapply(seq_along(rescList1), function(i)
{
attachMeans(normalized.counts=normExprData, design.matrix=desMat,
factor.column="gcondition", contrast.name=names(rescList1)[i],
de.results=rescList1[[i]])
})
names(rescList1) <- names
A volcano plot of differential expressed genes.
res.o.map1 <- convertGenesViaMouseDb(gene.list=rownames(rescList1[[1]]),
fromType="ENTREZID")
res.o <- attachGeneColumnToDf(mainDf=rescList1[[1]],
genesMap=res.o.map1,
rowNamesIdentifier="ENTREZID",
mapFromIdentifier="ENTREZID",
mapToIdentifier="SYMBOL")
WriteDataFrameAsTsv(data.frame.to.save=res.o,
file.name.path=paste0(names(rescList1)[1], "_edgeR"))
vp <- luciaVolcanoPlot(res.o, positive.controls.df=NULL, prefix=names(rescList1)[1],
threshold=padj.thr)
ggplotly(vp)
de <- sum(res.o$FDR < padj.thr)
nde <- sum(res.o$FDR >= padj.thr)
detable <- cbind(de,nde)
rownames(detable) <- names(rescList1)[1]
ddetable <- detable
tot.ctrls <- dim(sd.pos.ctrls)[1]
idx.pc <- which(tolower(res.o$gene) %in% tolower(sd.pos.ctrls[,1]))
tot.pc.de <- sum(res.o$FDR[idx.pc] < padj.thr)
tot.pc.nde <- length(idx.pc) - tot.pc.de
pos.df <- cbind(tot.ctrls, tot.pc.de, tot.pc.nde)
colnames(pos.df) <- c("total_p.ctrl", "p.ctrl_de_mapped",
"p.ctrl_notde_mapped")
rownames(pos.df) <- names(rescList1)[1]
kowthc5 <- res.o[which(res.o$FDR < padj.thr),]
kowthc5.sign.genes.entrez <- rownames(res.o)[which(res.o$FDR < venn.padgj.thr)]
A volcano plot of differential expressed genes.
rs2.o.map <- convertGenesViaMouseDb(gene.list=rownames(rescList1[[2]]),
fromType="ENTREZID")
res.rs2.o <- attachGeneColumnToDf(mainDf=rescList1[[2]],
genesMap=rs2.o.map,
rowNamesIdentifier="ENTREZID",
mapFromIdentifier="ENTREZID",
mapToIdentifier="SYMBOL")
WriteDataFrameAsTsv(data.frame.to.save=res.rs2.o,
file.name.path=paste0(names(rescList1)[2], "_edgeR"))
vp <- luciaVolcanoPlot(res.rs2.o, positive.controls.df=sd.pos.ctrls,
prefix=names(rescList1)[2],
threshold=padj.thr)
ggplotly(vp)
de <- sum(res.rs2.o$FDR < padj.thr)
nde <- sum(res.rs2.o$FDR >= padj.thr)
detable <- cbind(de,nde)
rownames(detable) <- names(rescList1)[2]
ddetable <- rbind(ddetable, detable)
kowtsd5 <- res.rs2.o[which(res.rs2.o$FDR < padj.thr),]
kowtsd5.sign.genes.entrez <- rownames(res.rs2.o)[which(res.rs2.o$FDR < venn.padgj.thr)]
A volcano plot of differential expressed genes.
res.o.map <- convertGenesViaMouseDb(gene.list=rownames(rescList1[[3]]),
fromType="ENTREZID")
res.o <- attachGeneColumnToDf(mainDf=rescList1[[3]],
genesMap=res.o.map,
rowNamesIdentifier="ENTREZID",
mapFromIdentifier="ENTREZID",
mapToIdentifier="SYMBOL")
WriteDataFrameAsTsv(data.frame.to.save=res.o,
file.name.path=paste0(names(rescList1)[3], "_edgeR"))
vp <- luciaVolcanoPlot(res.o, positive.controls.df=NULL,
prefix=names(rescList1)[3], threshold=padj.thr)
ggplotly(vp)
de <- sum(res.o$FDR < padj.thr)
nde <- sum(res.o$FDR >= padj.thr)
detable <- cbind(de,nde)
rownames(detable) <- names(rescList1)[3]
ddetable <- rbind(ddetable, detable)
kowtsrs2 <- res.o[which(res.o$FDR < padj.thr),]
kowtrs2.sign.genes.entrez <- rownames(res.o)[which(res.o$FDR < venn.padgj.thr)]
We present a summarization of the results. The first table is a summarization on how many genes are Differentially Expressed. The second table explains on the first column how many positive controls we have, on the second column how many positive controls have been identified over the differentially expressed genes, and, finally, on the third column how many positive controls have beed identified on the NOT differentially expressed genes.
ddetable
## de nde
## S3HC5 - WTHC5 28 14467
## S3SD5 - WTSD5 55 14440
## S3RS2 - WTRS2 23 14472
pos.df
## total_p.ctrl p.ctrl_de_mapped p.ctrl_notde_mapped
## S3HC5 - WTHC5 579 3 553
We take the results of two contrasts.
Knock Out Sleed Deprivation VS Wild Type Sleep Deprivation and Knock Out Home Cage control VS Wild Type Home Cage Controls . And plot the results in a Venn Diagram
source("../../R/venn2.R")
source("../../R/venn3.R")
gene.map <- convertGenesViaMouseDb(gene.list=rownames(normExprData),
fromType="ENTREZID", toType="SYMBOL")
venn <- Venn2de(x=kowthc5.sign.genes.entrez, y=kowtsd5.sign.genes.entrez,
label1="S3HC5_WTHC5", label2="S3SD5_WTSD5",
title="S3HC5_WTHC5 venn S3SD5_WTSD5", plot.dir="./",
conversion.map=gene.map)
We take the results of three contrasts.
Shank3 Sleed Deprivation VS Wild Type Sleep Deprivation and Shank3 Home Cage Control VS Wild Type Home Cage Controls and Shank3 Home Recovery Sleep VS Wild Type Recovery Sleep. And plot the results in a Venn Diagram
venn3 <- Venn3de(x=kowthc5.sign.genes.entrez, y=kowtsd5.sign.genes.entrez,
z=kowtrs2.sign.genes.entrez,
label1="S3HC5_WTHC5", label2="S3SD5_WTSD5", label3="S3RS2_WTRS2",
title="S3HC5_WTHC5 - S3SD5_WTSD5 - S3RS2_WTRS2", enrich.lists.flag=FALSE,
plot.dir="./",
conversion.map=gene.map)
An heatmap of the union of the genes identified.
source("../../R/heatmapFunctions.R")
de.genes.entr <- union(rownames(venn$int), rownames(venn$XnoY))
de.genes.entr <- union(de.genes.entr, rownames(venn$YnoX))
de.genes.entr <- union(kowthc5.sign.genes.entrez, kowtsd5.sign.genes.entrez)
de.genes.entr <- union(de.genes.entr, kowtrs2.sign.genes.entrez)
gene.map <- convertGenesViaMouseDb(gene.list=de.genes.entr,
fromType="ENTREZID")
de.genes.symb <- attachGeneColumnToDf(as.data.frame(de.genes.entr,
row.names=de.genes.entr),
genesMap=gene.map,
rowNamesIdentifier="ENTREZID",
mapFromIdentifier="ENTREZID",
mapToIdentifier="SYMBOL")
# de.genes.symb[which(is.na(de.genes.symb$gene)),]
de.genes.symb$gene[which(de.genes.symb$de.genes.entr=="100039826")] <- "Gm2444" ## not annotated in ncbi
de.genes.symb$gene[which(de.genes.symb$de.genes.entr=="210541")] <- "Entrez:210541" ## not annotated in ncbi
de.genes.counts <- normExprData[match(de.genes.symb$de.genes.entr, rownames(normExprData)),]
rownames(de.genes.counts) <- de.genes.symb$gene
de.gene.means <- computeGeneMeansOverGroups(counts=de.genes.counts,
design=designMatrix, groupColumn="gcondition")
library(gplots)
library(clusterExperiment)
color.palette = clusterExperiment::seqPal3#c("black", "yellow")
pal <- colorRampPalette(color.palette)(n = 1000)
library(pheatmap)
filter2 <- rowMeans(de.gene.means)>0
filter <- apply(de.gene.means, 1, function(x) log(x[4]/x[3]) * log(x[2]/x[1]) < 0)
filter[is.na(filter)] <- FALSE
with k=5 the samples S3RS2 seems to be mixed with the S3SD5 ones.
ph1 <- pheatmap(log(de.genes.counts[filter2,]+1), cluster_cols=TRUE, scale="row",
color=pal, border_color=NA, fontsize_row=5)
ph1 <- pheatmap(log(de.genes.counts[filter2,]+1), cluster_cols=TRUE, scale="row",
color=pal, border_color=NA, fontsize_row=10, kmeans_k=5)
clusterized.genes <- as.data.frame(ph1$kmeans$cluster)
gene.map <- convertGenesViaMouseDb(gene.list=rownames(clusterized.genes), fromType="SYMBOL")
converted.clusterized.gens <- attachGeneColumnToDf(mainDf=clusterized.genes, genesMap=gene.map,
rowNamesIdentifier="SYMBOL", mapFromIdentifier="SYMBOL", mapToIdentifier="ENTREZID")
converted.clusterized.gens$gene[which(rownames(converted.clusterized.gens)=="Gm2444")] <- "100039826" ## not annotated in ncbi
converted.clusterized.gens$gene[which(rownames(converted.clusterized.gens)=="Entrez:210541")] <- "210541" ## not annotated in ncbi
converted.clusterized.gens <- converted.clusterized.gens[order(converted.clusterized.gens$`ph1$kmeans$cluster`),]
WriteDataFrameAsTsv(data.frame.to.save=converted.clusterized.gens, file.name.path="clustered_genes_by_kmeans5")
Using k=6 we maintain the clustering of S3RS2 samples and S3SD5 in two different clusters of columns.
ph1 <- pheatmap(log(de.genes.counts[filter2,]+1), cluster_cols=TRUE, scale="row",
color=pal, border_color=NA, fontsize_row=5)
ph1 <- pheatmap(log(de.genes.counts[filter2,]+1), cluster_cols=TRUE, scale="row",
color=pal, border_color=NA, fontsize_row=10, kmeans_k=6)
clusterized.genes <- as.data.frame(ph1$kmeans$cluster)
gene.map <- convertGenesViaMouseDb(gene.list=rownames(clusterized.genes), fromType="SYMBOL")
converted.clusterized.gens <- attachGeneColumnToDf(mainDf=clusterized.genes, genesMap=gene.map,
rowNamesIdentifier="SYMBOL", mapFromIdentifier="SYMBOL", mapToIdentifier="ENTREZID")
converted.clusterized.gens$gene[which(rownames(converted.clusterized.gens)=="Gm2444")] <- "100039826" ## not annotated in ncbi
converted.clusterized.gens$gene[which(rownames(converted.clusterized.gens)=="Entrez:210541")] <- "210541" ## not annotated in ncbi
converted.clusterized.gens <- converted.clusterized.gens[order(converted.clusterized.gens$`ph1$kmeans$cluster`),]
WriteDataFrameAsTsv(data.frame.to.save=converted.clusterized.gens, file.name.path="clustered_genes_by_kmeans6")